# US 2021/0174027 A1 · Assertion Detection in Multi-Labelled Clinical Text using Scope Localization

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0174027 A1
- Document ID: US20210174027A1
- Status: Published application
- Kind: A1
- Application number: 16/949,838
- Filed: 2020-11-17
- Published: 2021-06-10
- Earliest claimed: 2019-12-10
- Assignees: Siemens Healthcare GmbH
- Applicants: Siemens Healthcare GmbH
- Inventors: Rajeev Bhatt Ambati; Oladimeji Farri; Ramya Vunikili
- Web page: https://patentgrep.com/patent/US20210174027A1

## CPC

- G06F 40/30: Semantic analysis
- G06F 40/279: Recognition of textual entities
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G16H 30/20: for handling medical images, e.g. DICOM, HL7 or PACS
- G16H 30/40: for processing medical images, e.g. editing
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G16H 70/60: relating to pathologies
- A61B 6/032: {Transmission computed tomography [CT]}
- A61B 6/503: {for diagnosis of the heart}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)

## Abstract

For assertion detection from clinical text in a medical system, a model, such as a neural network, is trained to operate on multi-labeled clinical text. Using multi-task learning, both the scope and the class losses are minimized. As a result, a machine learning model can predict both the scope and class of clinical text for a patient where the clinical text is not limited to one class or a particular length.

## Claims

### Claim 1

A method for assertion detection from clinical text in a medical system, the method comprising:
inputting the clinical text to a machine learning model;
identifying both a scope as a word group box and an assertion class for the word group box from the clinical text, the machine learning model identifying both the word group box and the assertion class in response to the inputting; and
generating an image showing words of the word group box and the assertion class for the words.

### Claim 2 (depends on 1)

The method of claim 1 wherein inputting the clinical text comprises inputting a radiology report for a patient.

### Claim 3 (depends on 1)

The method of claim 1 wherein inputting comprises inputting to the machine learning model, the model comprising a convolutional neural network.

### Claim 4 (depends on 1)

The method of claim 1 wherein inputting comprises inputting to the machine learning model, the model having been trained as a multi-task model with a combination of an objective function for the scope and an objective function for the assertion class.

### Claim 5 (depends on 4)

The method of claim 4 wherein the objective function for the scope comprised a mean square error between predicted and ground-truth intersection over union.

### Claim 6 (depends on 4)

The method of claim 4 wherein the objective function for the assertion class comprised a cross entropy loss.

### Claim 7 (depends on 4)

The method of claim 4 wherein the combination comprises a sum.

### Claim 8 (depends on 1)

The method of claim 1 wherein identifying the scope comprises identifying a plurality of possible word groups by the machine learning model and identifying the scope with a non-max suppression of the possible word groups.

### Claim 9 (depends on 1)

The method of claim 1 wherein identifying the scope comprises identifying the scope as a word group forming the word group box of a same class.

### Claim 10 (depends on 1)

The method of claim 1 wherein identifying the assertion class comprises identifying the assertion class as one of present, absent, conditional, hypothetical, possible, or associated with someone else.

### Claim 11 (depends on 1)

The method of claim 1 wherein identifying comprises identifying both the scope and the assertion class as a single stage, end-to-end operation of the machine learning model.

### Claim 12

A system for assertion detection from clinical text, the system comprising:
a memory configured to store the clinical text for a patient;
a processor configured to separate the clinical text into multiple assertions, the clinical text separated by a machine learning model configured to localize a scope in the clinical text for each of the assertions and a class for each of the assertions, the machine learning model having been trained with multi-labelled data as a multiple-task model; and
a display configured to indicate the scope and class for each assertion.

### Claim 13 (depends on 12)

The system of claim 12 wherein the machine learning model comprises a convolutional neural network.

### Claim 14 (depends on 12)

The system of claim 12 wherein the machine learning model was trained as the multi-task model with an objective function for the scope and an objective function for the class.

### Claim 15 (depends on 12)

The system of claim 12 wherein the processor is configured to select a localization of the scope output by the machine learning model using non-max suppression.

### Claim 16

A method for training a machine learning model for assertion detection, the method comprising:
machine training a neural network to both detect class and scope for each instance in clinical text, the machine-training using multi-label text as ground truth where the multi-labels include the class and the scope;
storing the trained neural network.

### Claim 17 (depends on 16)

The method of claim 16 wherein machine training comprises performing multi-task learning with a combination loss of a first objective function for the class and a second objective function for the scope being minimized.

### Claim 18 (depends on 16)

The method of claim 16 wherein machine training the neural network comprises machine training a convolutional neural network.

### Claim 19 (depends on 16)

The method of claim 16 further comprising selecting from possible word groupings as the scope with non-max suppression.

### Claim 20 (depends on 16)

The method of claim 16 wherein machine training comprises machine training as a single stage, end-to-end solution for both the scope and the class.
